[thclark]

A reader following the getting-started guide on a Linux laptop reported that the single-point example (alpha 5) did not converge while lower alphas did, and that only a small part of the NACA 4412 polar was valid.

Findings

Neither depends on the host. macOS, Linux arm64 and Linux x86_64 (Ubuntu 24.04 containers) give identical panels, identical results and a byte-identical polar JSON. Both problems are in the docs:

• Single point. NACA 0012 at alpha 5, Re 1e6 does not converge from a fresh start, and more iterations don't help (tried 30, 50 and 100). It is an isolated point: 4.5 and 5.5 converge. The pristine DP reference XFOIL fails at the same point on the same panels (VISCAL: Convergence failed), and at 4.5 it converges in 8 iterations, as yFoil does. "Your first analysis" had sidestepped this with Re 1e9.
• 4412 polar. The Polars page never said how naca4412.json was made. The only command producing it earlier in the guide, on the Geometry page, makes a sharp-TE, 200-node section. On that section the sweep fails from alpha 1.5 and halts after four failures at alpha 3, giving 13/41 valid. XFOIL does exactly the same (last converged alpha 1.0, CL 0.49137 in both). The 160-node blunt-TE section the docs figure was made from gives 41/41 valid.

Changes

• Single-point examples (Your first analysis; Analyse a single point) now use alpha 4 at Re 1e6, which converges in 7 iterations.
• The Polars page now generates the section: yfoil geometry naca 4412 -n 160 -o naca4412.json.
• The Geometry page's 4412 example is now 160 nodes with a blunt TE. --sharp is still documented further down that page.
• scripts/guide-figures.sh runs the new commands. naca0012_a5.fig becomes naca0012_a4.fig, and the other seven figures regenerate byte-identical.

scripts/docs-check.py passes.